Getting started
Rules Engine Widgets
OpenAI Agent Node
39 min
openai agent node table of contents what is the openai agent node? quick comparison main settings (4 sections) detailed guide for each setting real world examples how to use it common questions troubleshooting what is the openai agent node? the openai agent node is a tool that lets you add an ai assistant powered by openai (like chatgpt) to your workflow think of it like this you give the agent a job description (instructions) you give it a task to do (user input) the agent does the work and gives you results what it can do answer questions and provide analysis extract information from text generate content (emails, reports, summaries) solve problems step by step work with data and make decisions call other tools and services quick comparison openai agent vs universal agent feature openai agent universal agent ai models only openai (gpt 4, gpt 5, etc) multiple (openai, claude, deepseek, gemini) ease of use simpler, less options more options, more complex best for if you only want openai if you want choices speed faster (optimized for openai) slightly slower choose openai agent if you want to use openai and keep setup simple choose universal agent if you want to try different ai providers main settings (4 sections) when you open an openai agent node, you see 4 main areas to configure detailed guide for each setting section 1 agent setup 1 1 agent name what it is a label/name for your agent why it matters helps you identify this agent in your workflow how to fill it give it a descriptive name use clear, simple names no special characters needed good names customer support helper sales data analyzer email writer review classifier product description generator bad names agent1 temp test my agent node abc 1 2 instructions what it is what you tell the agent to do and how to behave why it matters better instructions = better results the agent will follow what you write here exactly how to fill it tell the agent what job it has how it should work what rules to follow example support agent you are a customer support specialist your job is to help customers solve their problems quickly and kindly when a customer contacts you • read their message carefully and understand what's wrong • be friendly and show you care about their problem • ask questions if something is unclear • give step by step solutions that are easy to follow • if you can't fix it after trying 3 times, offer to connect them to a human agent do not • share passwords or sensitive account information • make promises about refunds without checking company policy • be rude or dismissive • give up after one attempt pro tips the better you describe the job, the better the agent works be specific about what you want say what the agent should not do explain the tone/style you want (professional, casual, etc ) 1 3 user input/prompt what it is the actual task or question you want the agent to work on why it matters this is what the agent actually processes how to fill it option a type it directly "summarize this article \[paste article text]" option b pull from previous step "analyze this customer feedback {{dataflow\ previous()}}" this automatically uses the output from the previous step in your workflow option c mix both "analyze this data {{dataflow\ previous()}}focus on profitability and growth trends" available shortcuts {{dataflow\ previous()}} = use output from the last step {{dataflow\ current()}} = use current workflow data {{dataflow\ getbynodeid('node name')}} = use output from a specific step examples example 1 classification input "tell me if this review is positive, negative, or neutral {{dataflow\ previous()}}" what happens agent gets the review from previous step and classifies it example 2 summarization input "summarize this in 100 words {{dataflow\ getbynodeid('document fetcher')}}" what happens agent gets document and creates short summary example 3 analysis input "compare these two datasets\ dataset a {{dataflow\ getbynodeid('sales data')}} dataset b {{dataflow\ getbynodeid('forecast data')}} tell me the top 3 differences" what happens agent compares and lists differences 1 4 authentication (api key) what it is your openai password/credential to access the service why it matters this is like your account login the agent needs it to access openai how to fill it option a use stored credential 1\ click "use stored credential" button 2\ select your openai api key from the list 3\ done! (key is secure) benefits more secure (key is encrypted) easier to manage (update once, works everywhere) safer for team workflows can't accidentally expose the key option b enter api key directly 1\ copy your openai api key (starts with "sk ") 2\ paste it in the box 3\ done! (but not recommended for production) use this only for quick testing personal projects learning/experimentation status indicator symbol meaning what to do 🟢 green checkmark api key is valid and working you're good to go! 🔴 red x or warning api key is missing or incorrect check your key or add it security tips always use "stored credential" if possible never share your api key with others don't put api key in code or workflows you share change your api key if you think someone saw it section 2 model & behavior ( need your review on this team) 2 1 model selection what it is which ai brain to use (different strengths and speeds) why it matters different models are good at different things choosing the right one saves money and gets better results available models model name what it's good at speed cost gpt 5 1 best at everything, complex analysis slower most expensive gpt 5 very smart, good for difficult tasks slow expensive gpt 4 1 great balance of smart & fast fast moderate gpt 4o good with text and images fast moderate gpt 4 reliable, works well fast moderate o1 deep thinking, solving hard problems slowest most expensive gpt 3 5 turbo quick and cheap, simple tasks fastest cheapest how to choose use gpt 5 1/gpt 5 if you need the best results task is very complex cost is not a concern use gpt 4 1/gpt 4 if good balance of price and quality most common choice works for almost everything use gpt 3 5 turbo if task is simple you want to save money speed is important example classifying emails, simple summaries use o1 if problem is very complex need deep reasoning time is not an issue example selections textcustomer support → gpt 4 1 (reliable, good speed)data analysis → gpt 5 (needs to be smart) email classification → gpt 3 5 turbo (simple task, save money) code review → o1 (deep analysis needed) content writing → gpt 4 1 (balanced) 2 2 agent type what it is how the agent works in your workflow two options type 1 orchestrator (default) agent works by itself takes input and gives output other nodes use its output most common choice flow \[previous step] → \[orchestrator agent] → \[next step] type 2 agent as tool another agent can call this agent creates sub agent system useful for specialization for advanced workflows \[main manager agent] ↓ calls ├─→ \[research agent] ├─→ \[writing agent] └─→ \[fact checking agent] when to use which use orchestrator when agent works alone output goes to next step simple workflow use agent as tool when building advanced multi agent system one agent manages multiple sub agents each agent has special expertise 2 3 temperature what it is controls how creative vs consistent the agent is think of it as a "creativity slider" from 0 to 2 value behavior use for example task 0 0 3 robotic, same answer every time data extraction, facts, classifications extract email address from this text 0 4 0 6 mostly consistent, slight variations support tickets, analysis analyze this data and find trends 0 7 default balanced most general tasks help this customer with their issue 0 8 1 2 more creative, varied responses content writing, brainstorming write a creative product description 1 3 2 0 very creative, unpredictable experimental, creative writing generate 10 unique marketing taglines 2 4 max tokens what it is the maximum length of the agent's response tokens are like "ai words " 1 token ≈ 4 characters ≈ ¾ of a word example "hello world" = about 2 tokens more tokens = longer responses = higher cost token guide tokens roughly equals good for example use 100 75 words very short answers "yes/no", "positive/negative", one word classifications 256 200 words short answers brief summaries, quick classifications 500 375 words medium paragraph email responses, short explanations 1,000 750 words multiple paragraphs default most tasks 2,000 1,500 words long response detailed analysis, reports 4,000 3,000 words very long response comprehensive reports, articles default 1,000 tokens (usually enough) when to increase need longer answers detailed report required complex analysis needed when to decrease want quick, short answers saving on api cost brief response needed example sentiment analysis 100 tokens (just positive/negative/neutral) email summary 500 tokens (brief summary) product analysis 2,000 tokens (detailed breakdown) section 3 tools & integrations 3 1 what are tools? tools are extra helpers the agent can use to do more like giving the agent superpowers can search the internet can call other services can process data can do calculations 3 2 available tools system tools (built in) tool what it does use case web search searches google/bing for current info find latest news, prices, trends http request calls any website/api get data from other services json processor works with data/code clean up messy data text processor advanced text work extract emails, phone numbers math calculator does math calculate percentages, sums text summarizer shortens text with ai summarize long documents 3 3 how to enable tools steps click on "🛠️ tools & integrations" section see list of available tools check the boxes for tools you want agent can now use those tools tool examples in action example 1 web search what it does agent can search the internet for current information enable when need real time data (weather, stock prices, news) looking up facts that might change researching current events example task instructions "you are a research assistant answer questions using current information " user input "what are the top 3 ai trends in 2025?" tools enabled ✅ web search what happens agent searches google, reads results, and summarizes top trends example 2 http request what it does agent can make api calls to other services enable when need to fetch data from an external api check status in another system send data to another service example task instructions "you help customers check their order status "user input "check order #12345" tools enabled ✅ http request what happens agent calls your order api, gets status, and replies to customer example 3 math calculator what it does agent can do precise calculations enable when need accurate math (not approximations) calculating prices, discounts, taxes working with numbers example task instructions "you calculate discounts for customers " user input "apply 20% off to a $150 purchase, then add 8% tax" tools enabled ✅ math calculator what happens agent does ($150 0 80) 1 08 = $129 60 example 4 json processor what it does agent can work with json data (common data format) enable when processing api responses cleaning up messy data transforming data structures example task instructions "extract customer emails from this json data " user input \[json data with customer records] tools enabled ✅ json processor what happens agent parses json and extracts all email addresses section 4 advanced features 4 1 error handling what it is what the agent does when something goes wrong three options option 1 graceful (continue with warnings) default if something fails, agent keeps going tells you about the problem gives best answer it can best for most situations option 2 strict (stop on error) if something fails, everything stops makes sure nothing wrong gets through best for critical operations (payments, medical info) option 3 continue (ignore errors) if something fails, agent ignores it keeps going without telling you best for non important tasks, experiments example 4 2 parallel tool execution what it is should the agent use multiple tools at the same time or one after another? two options enabled (parallel) agent uses many tools simultaneously faster results better for independent tasks disabled (sequential) agent uses one tool, then the next slower but more controlled better when tools depend on each other when to enable getting data from multiple sources at once multiple independent searches parallel calculations when to disable one tool needs output from another 4 3 agent delegation what it is can this agent ask other agents to help? simple analogy like a manager assigning work to team members how it works main agent "task coordinator"├─ assigns to agent 1 "extract data" ├─ assigns to agent 2 "analyze data" └─ assigns to agent 3 "write report" main agent then combines all results when to use complex tasks needing different specialties multiple sub agents for different areas building agent teams how to set it up click "agent delegation" section click "add targets" select which agents can be called main agent can now use them 4 4 conversation memory what it is can the agent remember past conversations? why it matters agent understands context conversations feel natural avoids repeating information more helpful interactions when to enable ✅ customer support (remember customer history) ✅ multi turn conversations ✅ ongoing projects when to disable ❌ one time tasks ❌ privacy sensitive data ❌ don't need context how to enable click "💾conversation memory" check "enable conversation memory" memory is now active example customer support agent goal help customers with their problems configuration agent name customer support agent instructions "you are a customer support specialist your job is to help customers solve problems when a customer contacts you 1\ be friendly and empathetic 2\ ask questions to understand the issue 3\ give clear, step by step solutions 4\ if you can't solve it after 3 attempts, offer to connect them to a human tone professional but friendly language simple, non technical" user input "customer issue {{dataflow\ previous()}}" model gpt 4 1 temperature 0 7 maxtokens 1000 tools enabled ✅ web search (to find solutions) ✅ http request (to check account status) error handling graceful (continue even if one tool fails) memory enabled (remember customer history) workflow